EMWaveNet: Physically Explainable Neural Network Based on Electromagnetic Propagation for SAR Target Recognition

TL;DR

This paper tackles the lack of transparency in SAR target recognition by introducing EMWaveNet, a physically explainable, complex-valued neural network grounded in electromagnetic wave propagation. It leverages a modulation-layer architecture and a novel loss L_{SNR} to steer energy to category regions, yielding high accuracy while preserving clear physical interpretations of all learnable parameters. The approach demonstrates strong de-overlapping capabilities and robustness to forest interference and other complex perturbations, including end-to-end SAR-ATR in interference scenes. These results suggest that integrating physical propagation models into complex-valued DL can enhance explainability, robustness, and practical deployment of SAR recognition systems in realistic environments.

Abstract

Deep learning technologies have significantly improved performance in the field of synthetic aperture radar (SAR) image target recognition compared to traditional methods. However, the inherent ``black box" property of deep learning models leads to a lack of transparency in decision-making processes, making them difficult to be widespread applied in practice. To tackle this issue, this study proposes a physically explainable framework for complex-valued SAR image recognition, designed based on the physical process of microwave propagation. This framework utilizes complex-valued SAR data to explore the amplitude and phase information and its intrinsic physical properties. The network architecture is fully parameterized, with all learnable parameters endowed with clear physical meanings. Experiments on both the complex-valued MSTAR dataset and a self-built Qilu-1 complex-valued dataset were conducted to validate the effectiveness of framework. The de-overlapping capability of EMWaveNet enables accurate recognition of overlapping target categories, whereas other models are nearly incapable of performing such recognition. Against 0dB forest background noise, it boasts a 20\% accuracy improvement over traditional neural networks. When targets are 60\% masked by noise, it still outperforms other models by 9\%. An end-to-end complex-valued synthetic aperture radar automatic target recognition (SAR-ATR) algorithm is constructed to perform recognition tasks in interference SAR scenarios. The results demonstrate that the proposed method possesses a strong physical decision logic, high physical explainability and robustness, as well as excellent de-aliasing capabilities. Finally, a perspective on future applications is provided.

Figures (15)

• Figure 1: (a) Difficulties in SAR image classification, (b) Traditional CNNs are characterized by opaque internal mechanisms and uninterpretable decision-making processes, while, our method is designed for complex-valued SAR images. The parameters learned possess clear physical significance, thereby enhancing the explainability of the network.
• Figure 2: The overall of EMWaveNet. The network consists of three parts. The input complex-valued SAR image, several modulation layers in the middle and the final classification detection layer.
• Figure 3: CNN vs. EMWaveNet
• Figure 4: Comparison of two network structures. (a) Traditional Convolutional Neural Network: The architecture consists of layers of neurons where each neuron in a given layer is connected to the neurons in the previous and subsequent layers. These connections allow the network to learn spatial hierarchies of features through convolution and pooling operations. (b) EMWaveNet: This network is designed to model the propagation of electromagnetic waves in a vacuum, where the process of wave propagation is modulated by the network’s parameters. EMWaveNet incorporates the principles of electromagnetic wave theory to enhance the network’s ability to capture complex wave behavior in SAR imaging.
• Figure 5: The detail of modulation layer. Each neuron in the modulation layer has a set of learnable parameters: amplitude and phase.